Large difference in pLI from v2.1.1 to v4.1.0

Hi there,

I’ve noticed that there’s a striking difference in pLI for MAPT between v2.1.1 (pLI=0.05) and v4.1.0 (pLI=1). Any ideas what might be driving this difference? The o/e and LOEUF are pretty similar. I spot checked a few other genes and didn’t see much of a difference between versions.


Hello and thanks for posting! Could you send the links that you’re using query this? When I check MAPT on our browser for v4.1 I see pLI==0 .


I’m so sorry, I mis-typed. I meant APP, not MAPT!


The assumption underlying pLI is that we can assign genes to three categories based on their sensitivity to loss-of-function (LoF) variation:

  • Tolerant (Null): Heterozygous and homozygous LoF variation in the gene are completely tolerated by natural selection
  • Recessive (Rec): Heterozygous LoF variation is tolerated but homozygous LoF variation is not
  • Haploinsufficient (LI): Intolerant to LoF variation (heterozygous or homozygous)

The values used to assign genes to these three categories in v2 were:

{"Null": 1.0, "Rec": 0.463, "LI": 0.089}

We updated these values for v4 to the following (based the new observed and expected variant counts seen in the larger dataset):

{"Null": 1.0, "Rec": 0.706, "LI": 0.207}

The observed to expected ratio (OE) of predicted LoF (pLoF) variation in APP in gnomAD v2 is 0.26, and the pLoF OE in v4 is 0.30. This means that in v4, the pLoF OE makes APP look more like it falls in the haploinsufficient category than it did in v2 (OE of 0.30 in v4 is close to LI OE of 0.207).

For more information about pLI, see the pLI section of our gene constraint help page.

Ah that makes sense, thanks very much!